Route a support request

One customer message can answer several questions for your application. This example combines D1’s three question types to help route a support request.

Define the questions

Save this body as prediction.json:

prediction.json

{
  "model": "default",
  "context": {
    "message": "My parcel arrived with a cracked mug. Please help.",
    "days_since_delivery": 1
  },
  "questions": [
    {
      "id": "needs_help",
      "type": "binary",
      "prompt": "Does the customer need support?"
    },
    {
      "id": "team",
      "type": "choice",
      "prompt": "Which team should handle this message?",
      "options": [
        { "id": "support", "description": "Customer support" },
        { "id": "sales", "description": "Sales" }
      ]
    },
    {
      "id": "urgency",
      "type": "score",
      "prompt": "How urgent is the issue?",
      "levels": ["Low urgency", "Moderate urgency", "High urgency"]
    }
  ]
}

Send one request

Set D1_API_URL and D1_API_KEY as shown in the quickstart. Generate a fresh key once, then keep it with this request for any retries.

export D1_OPERATION_KEY="$(python3 -c 'import uuid; print(uuid.uuid4())')"

curl --fail-with-body --connect-timeout 5 --max-time 40 \
  "$D1_API_URL/v1/predictions" \
  -H "Authorization: Bearer $D1_API_KEY" \
  -H 'Content-Type: application/json' \
  -H "Idempotency-Key: $D1_OPERATION_KEY" \
  --data-binary @prediction.json

Use the answers

Answers arrive in the same order and retain the IDs you supplied.

Question	Read	Use it for
`needs_help`	`probability_true`	Decide whether to queue the message for support
`team`	`selected` and `options`	Route to a team and inspect the scores for each option
`urgency`	`value`	Rank messages on the three-level rubric, from 0 to 2

A score can be fractional. Choose your own routing thresholds using representative examples; model scores do not establish probability calibration.

Staging answers are simulated. Use them to test request handling and routing code, then evaluate a real model before making decisions from its scores.

See Create a prediction for the full response schema, or reliable retries to handle an interrupted request.

​Define the questions

​Send one request

​Use the answers

Define the questions

Send one request

Use the answers